Research Collaboration Discovery through Neo4j Knowledge Graph
Keller Smith, Fang Cherry Liu, Deepa Phanish, H Chen, R Chen, Didier Contis · 2024
Knowledge Graphs (KGs) provide the unique opportunity to organize data with a strong focus on relationships. Neo4j, a graph database service, stores nodes and relationships rather than tables, thus presenting a scalable and flexible solution to manage KGs from the storage to the user interface layer. Georgia Institute of Technology (GT) has a strong need to understand its researchers’ collaboration effort within and outside the institution. We have, therefore, created a knowledge graph on publications by GT researchers using Neo4j. It is part of a pilot project on a centralized platform to support Enterprise KGs across the institute. The current graph database in production contains 471,396 nodes and 2,155,302 relationships. The nodes include GT authors, their co-authors and institutions, with GT works. The relationships reflect where an author has worked, if they currently work at GT, what papers they have authored, and what other institutions they have collaborated with. We aim to build a pilot centralized platform to support Enterprise Knowledge Graphs (KGs), which is a complementary approach to existing enterprise analytics efforts at GT. We have built an end-to-end data pipeline to process data from OpenAlex, a free catalog of scholarly publications and related entities. For improved data processing speed, we have used GT Partnership for an Advanced Computing Environment (PACE) cluster to inject the OpenAlex data to the Neo4j database while integrating all Georgia Tech’s work data. The total data processing time reduced from one day to a couple of hours, in comparison to serial processing. We have also improved OpenAlex’s data accuracy, as we encountered some issues with the data. For example, many authors have multiple nodes in OpenAlex, and some entities have incorrect connections to other entities. Additionally, we have integrated different data sources (e.g. Scopus, Georgia Tech LDAP, ORCID) to improve the data integrity and introduce new, relevant data. Finally, we used the Neo4j Bloom visualization tool to gain insight into external research collaborations.